Wednesday, May 24, 2017
After thinking about it overnight, I decided to restart the recorder-based performance tool from scratch. The original design was trying to reuse the exising printf-like RECORD statements. But those are not intended with the same purpose in mind at all. They are intended to provide post-mortem crash or event analysis. So the data you have in there may not even be relevant as a measurement point.
I still think it makes sense to put that in the recorder module, since this will reuse some of the technology, most notably the ring buffers. But the configuration will be designed to be remote-accessible over shared memory.